GPU-based methods for exploring parabolic partial differential equations (Q533522)

Numerical solutions to parabolic partial differential equations (PDEs) are important not only for the reason of obtaining approximate solutions but also play an important role in the investigation of the properties of the exact solutions to the parabolic PDE's. The disadvantage of this approach is that the numerical methods are relatively slow on central processing units (CPU). That is why it is quite difficult to obtain a rapid feedback about the solution over the time. The way to overcome this problem in recent days is to use graphics processing units (GPU), which present an efficient hardware for massively parallel computations and offer the user rapid feedback. The authors show how to map the mathematical algorithms onto GPU and mention the main differences between the numerical solution obtained on a CPU compared to that obtained on a GPU. For this purpose a nonlinear parabolic PDE that arises in combustion and blow up in finite time is investigated. The numerical solution is based on a finite difference method of Crank-Nicolson type. Several experiments for this problem for 1D up to 3D space dimensions are computed and the main features how to implement the problem on a GPU are described.